1. Identification and control of rapidly time-varying systems. Using the multiple models, switching, and tuning methodology, an unknown rapidly time-varying system is identified and controlled.
2. Simultaneous identification. Multiple unknown nonlinear dynamical systems are identified simultaneously. The technique uses the multiple models, switching, and tuning methodology.
3. Adaptive control of nonlinear system. A class of unknown nonlinear systems is controlled using support vector regression.
4. Interference Cancellation. Using an array of sensors, the signal-of-interest is extracted despite the presence of stronger interference signals and that the arrival-directions are unknown and varying with time.
5. Detection and Tracking of dynamic obstacles for Autonomous robot navigation. Vision sensors are employed as the outward looking sensor for this application. Features are extracted to separate the background from the objects. Statistical estimation methods
are used to model the motion of the objects.
6. Vision based mapping and localization for autonomous rover navigation. Appearance based mapping is combined with visual odometry to generate metric and topological maps of the environment. Learning mechanisms are incorporated for terrain inclination
and relief profiles.
7. Embedded Computational architectures for real time applications. Co-processors and custom data path units are designed and interfaced with soft-core processors to accelerate the computationally intensive parts of computer vision algorithms used for mapping, localization and path planning. Task level and data level parallelism is employed to map tasks to the different processing elements. Systolic array structures are used for iterative computations. Systems are prototyped on state-of-art FPGAs for validation. Custom networks-on-chip are designed to overcome the bottleneck of shared bus protocols.
8. Assessment of Vascular Risk using Photoplethysmography. Photoplethysmographic (PPG) signals were used for the assessment of cardiovascular risk in human beings. An Auto-Regressive eXogenous input (ARX) linear parametric model was used to extract features that represent the circulatory system and a support vector machine (SVM) was used for classifying the signals based on the four data segment selection policies: best fit, three-best fit, ten-best fit and average-best fit.
9. An intelligent wearable device to recognize emotions in children with autism. This project is a step to bridge the gap between autistic children and the world and to give them a sense of acceptance in the society. It is a wearable device in the form of a wrist band with an array of sensors to capture the physiological parameters of subjects. These parameters are analysed to recognize different emotions like: happy, sad and involvement.